From: William Paulson <billpaulson@...>
Date: 2012-06-08T03:42:00+09:00
Subject: Re: Deadlock in large vm_xmalloc while doing backtrace




On Jun 5, 2012, at 5:11 PM, Peter Zotov <whitequark@whitequark.org> wrote:

> Bill Paulson писал 05.06.2012 22:52:
>> I work on a fairly large Ruby application that runs as a daemon,
>> accepting requests for test jobs and distributing them. It runs on Ruby
>> 1.9.1-p429 currently on Linux Redhat 5. The job scheduler ("foreman")
>> occasionally locks up in one site or another with a stack trace for the
>> main thread that looks like the following (many levels of stack 

While not completely certain, this bug appears to be due to using the 'caller' method inside an object finalizer. I've written a bug report on a simpler program (http://bugs.ruby-Lang.org/issues/6558) which I think is related. Removing 'caller()' from the finalizer in our real code seems to have stopped the problems (knocks on wood and crosses fingers.)

Bill Paulson